Classifying Multistate DNA Origami: An Automated Approach with Minimal Labeling and Confidence-Based Filtering.

Area of Science:

• Nanotechnology
• Biotechnology
• Computational Biology

Background:

• DNA origami nanostructures exhibit diverse conformational states crucial for nanoscale fabrication.
• Atomic Force Microscopy (AFM) is vital for structural validation but manual classification is time-consuming.
• Automated classification is needed to overcome bottlenecks in analyzing complex nanostructures.

Purpose of the Study:

• To introduce an automated classification framework for DNA origami nanostructures.
• To minimize user annotation while maintaining high classification accuracy.
• To enable efficient, high-throughput analysis of multistate DNA nanostructures.

Main Methods:

• Developed a confidence-based filter to identify uncertain or aggregated objects.
• Implemented iterative retraining with selective relabeling and data augmentation.
• Utilized a human-in-the-loop approach for refining the classifier with minimal new labels.

Main Results:

• The framework achieved high accuracy with limited labeled data (20 images/class) and strong augmentation.
• Outperformed established few-shot and contrastive learning baselines on distinct DNA origami datasets.
• Significantly reduced analysis time compared to conventional manual methods.

Conclusions:

• The automated framework enables efficient, objective, and high-throughput classification of DNA nanostructures.
• Human-in-the-loop refinement effectively captures intraclass morphological diversity.
• This approach accelerates advancements in nanoscale fabrication and structural analysis.